How to Send Emails from an AI Agent (Step by Step)

A founder messaged me a few weeks ago. Their app had a security vulnerability. They needed to alert every user immediately, but their developer was offline and they had never touched an email dashboard before. They connected the AutoSend MCP server to Claude, described the situation, and had the campaign drafted, reviewed, and sent in about fifteen minutes. No dashboard. No developer needed.

That story captures something that has shifted in how people use email infrastructure. Agents are not just writing copy for you to paste somewhere. They are operating the whole thing.

This article covers three ways to send emails from an AI agent using AutoSend, when to use each one, and what the actual setup looks like.

The two jobs agents do with email

Before getting into the paths, it helps to separate two distinct jobs.

The first is sending transactional emails: a user signs up, your code calls an API, one email goes to one person. This is code-driven and happens inside your application. The agent here is a coding agent helping you write and ship that integration.

The second is running campaigns and automations: drafting templates, building audience segments, setting up drip sequences, scheduling and sending to thousands of contacts. This is operational work that used to require a dashboard. Now an agent can do it in a conversation.

AutoSend handles both. The right integration path depends on which job you need.

Path 1: REST API (for transactional emails from code)

If you are building an application and you want it to send emails when things happen, you want the REST API. This is the path for password resets, order confirmations, verification emails, any email triggered by a user action inside your product.

The endpoint is:

POST https://api.autosend.com/v1/mails/send
Authorization: Bearer AS_your_key_here

A minimal request looks like this:

const res = await fetch('https://api.autosend.com/v1/mails/send', {
	method: 'POST',
	headers: {
		Authorization: 'Bearer AS_your_key_here',
		'Content-Type': 'application/json',
	},
	body: JSON.stringify({
		to: { email: user.email, name: user.name },
		from: { email: '[email protected]', name: 'Your App' },
		subject: 'Welcome to Your App',
		html: "<h1>Welcome, {{name}}!</h1><p>You're in.</p>",
		dynamicData: { name: user.name },
	}),
});

The dynamicData field handles Handlebars variables in your HTML. You can also pass a templateId instead of inline HTML if you have a saved template in AutoSend. The full request and response schema is in the send mail API reference.

When to use this path: Any time your application code needs to trigger an email in response to a user action. A coding agent like Claude Code, Cursor, or Copilot can write this integration for you without you explaining the AutoSend API at all, if you use the skill (covered in Path 3 below).

Path 2: MCP server (for campaigns and automations via natural language)

The AutoSend MCP server lets an AI assistant like Claude operate your email stack through conversation. This is the path for non-technical founders, marketers, and anyone who needs to move fast without opening a dashboard.

The marketer story I mentioned earlier is a real example of this. Someone on our customer's team wanted to build an onboarding automation: a welcome email, a follow-up three days later based on whether the user had activated, then a re-engagement branch for users who had not. They opened Claude, described the flow in plain language, and worked through it in conversation. Claude used the MCP server to create the templates, build the automation, and set up the branch logic. The marketer never wrote a line of code.

How to connect the MCP server

The AutoSend MCP server runs at https://mcp.autosend.com/. It uses OAuth, so there is nothing to install on your machine. You add the URL and authenticate once.

In Claude Desktop:

1. Open Claude Desktop and click Customize in the top left
2. Go to Connectors and click +
3. Select Add custom connector, name it AutoSend, enter https://mcp.autosend.com/
4. Click Connect and log in to your AutoSend account

In Cursor:

Add this to your ~/.cursor/mcp.json:

{
	"mcpServers": {
		"autosend": {
			"url": "https://mcp.autosend.com/",
			"type": "sse"
		}
	}
}

In Claude Code:

claude mcp add autosend --transport sse https://mcp.autosend.com/

Once connected, you can say things like:

• "Create a campaign announcing our new feature and send it to all active users"
• "Build a three-email onboarding automation that starts when someone joins the waitlist"
• "Duplicate last month's newsletter, update the hero copy, and schedule it for Tuesday at 9 AM"

Claude picks the right tools, checks your lists and sender details, and executes. The full docs for the MCP server are at docs.autosend.com/ai/mcp-server.

If you want a deeper look at what the MCP server can do tool by tool, I covered the full surface in our email MCP server guide.

What MCP is good for

Campaigns and automations are where MCP shines. The reason is that these tasks involve multiple steps, decisions, and data lookups (which list, which sender, which suppression group). An agent that can hold a conversation and fetch that context on your behalf removes a lot of friction.

MCP is also the right path when the person doing the work is not technical. A founder who can describe what they want in words should not need to learn a dashboard to ship an email.

Path 3: AutoSend Skill (for coding agents)

The AutoSend skill is a different kind of integration. It is not a server your agent talks to at runtime. It is a context file that gives your coding agent everything it needs to write correct AutoSend integration code, without you having to explain the API.

Install it with one command in your project root:

npx skills add https://github.com/autosendhq/skills --skill AutoSend

That command detects which coding agent you are using and puts the skill file in the right place. For Cursor it goes in .cursor/rules/autosend.mdc. For Claude Code it goes in .claude/CLAUDE.md. For Windsurf, .windsurfrules.

After that, you set your API key:

export AUTOSEND_API_KEY=AS_your_key_here

Then you just ask your agent to integrate email:

Send a welcome email to a new user using AutoSend. The user's email is in the user.email variable.

The agent generates the correct endpoint, authentication header, request body, and error handling without you needing to paste docs or describe the API. It already has that context from the skill file.

When to use this path: When you are building an application with a coding agent and you need transactional emails woven into your code. Lovable, Bolt, v0, and Replit all work with the skill. So does Claude Code, Cursor, Codex, and Windsurf.

Choosing the right path

Situation	Path
Building an app and need code that sends transactional emails	REST API + Skill
Non-technical founder or marketer needs to send a campaign now	MCP
Setting up an automation sequence in natural language	MCP
Coding agent is building an app and should handle email itself	Skill
You want to build your own agent that programmatically manages campaigns	REST API

The paths are not mutually exclusive. A lot of our customers use MCP for campaign and automation work, and the REST API (often written with help from a coding agent using the skill) for transactional emails inside their product.

What agents cannot do that a human should

One thing worth naming: agents are good at execution but they still need judgment calls from you.

Before you send a campaign to your entire list, a human should review it. Before an automation goes live, a human should verify the entry criteria. AutoSend's MCP server has a confirmation gate built into the send tools for exactly this reason. Claude will preview what it is about to send and ask you to confirm before anything goes out.

That is not a limitation. That is the right design. The agent handles the operational work. You handle the decisions that carry real consequences.

FAQ

Can an AI agent send transactional emails through the MCP server? The MCP server is designed for campaigns, automations, and template management, not one-off transactional sends. For transactional emails triggered by user actions in your application, use the REST API directly. The skill makes it easy for a coding agent to write that integration.

Does the AutoSend MCP server work with ChatGPT, Cursor, and other tools besides Claude? Yes. The MCP server works with Claude Desktop, Claude Code, Cursor, Copilot, Codex, Windsurf, Antigravity, ChatGPT, and Raycast. The setup steps vary slightly per tool. The docs cover each one.

What is the difference between the MCP server and the AutoSend skill? The MCP server lets an agent operate AutoSend at runtime through natural language. The skill is a context file that tells a coding agent how to use the AutoSend REST API when writing code. Use MCP when you want an agent to manage campaigns and automations. Use the skill when a coding agent is building an application that needs to send email.

Do I need an AutoSend account to use the MCP server? Yes. The MCP server authenticates via OAuth with your AutoSend account. You connect it once and stay authenticated. There is nothing to install on your machine.

Can I use the REST API and MCP server at the same time? Yes, and many customers do. The REST API handles transactional sends from your application code. The MCP server handles campaign and automation work. They operate on the same AutoSend account and the same sender reputation.

What happens to deliverability when an agent sends at scale? The same rules apply whether a human or an agent hits the send button. Your domain reputation, authentication setup, and list quality are what determine inbox placement. If you are new to sending at volume, read our domain warmup guide before your first large campaign.